Q&A is where you write down the exact question-and-answer pairs you want your agent to know, in your own words. Unlike Web Pages or Documents, which the agent has to read and interpret on its own, a Q&A pair tells it precisely what to say - which is why Q&A outranks every other training source your agent has.
That makes Q&A the tool you reach for in two situations: when the agent already got something wrong and you need to fix it fast, and when visitors keep asking the same handful of questions and you want a guaranteed-correct answer on file. You don't need to write a Q&A pair for everything your agent might be asked - crawled Web Pages, uploaded Documents, and the Video Transcript are built for broad coverage. Use Conversations when you want help building more Q&A from past chats (conversations themselves are not training data). Save hand-written Q&A for the answers you need to guarantee.
Why Q&A pairs come first
What it is
A Q&A pair is a question written the way a visitor might actually ask it, paired with the exact answer you want your agent to give back. You (or your team) write these directly - one at a time, or in bulk.
What it's best for
- Correcting an answer the agent got wrong when it relied on Web Pages, Documents, or the Video Transcript
- Locking in the right answer to a question you know visitors ask constantly
- Stating a policy, price, or fact in exact wording, with no room for the agent to paraphrase it incorrectly
How it works
When your agent decides how to answer a visitor, it checks Q&A first. If a visitor's question is a close enough match to a Q&A pair, the agent uses that answer instead of piecing one together from Web Pages, Video Transcript, or Documents. That priority is what makes Q&A the fastest way to override an answer you don't like - you don't have to hunt down and fix the underlying source material, you just add a Q&A pair that says what you actually want said, and it takes over from there.
Best practices
Treat Q&A as your override list, not your encyclopedia. Every time you see the agent give a wrong or off-tone answer in Chat Logs, write a Q&A pair to correct it - that's the single fastest way to improve accuracy, faster than editing a source document and waiting for it to be re-learned. Keep answers tight: a bloated Q&A answer is harder for the agent to blend naturally into a reply than a short, direct one.
Writing and importing Q&A pairs
What it is
You can add Q&A pairs two ways: type them straight into the dashboard one at a time, or import a batch at once from a file.
How it works
To write a pair by hand, open the Q&A tab under Training and add a question with its answer. To add many at once, use Upload Q&A: drop in a CSV, Excel (.xlsx), JSON, or NDJSON file, or paste JSON directly. If your file uses its own layout, Describe format lets you explain which columns or fields hold the questions and answers. This is the fastest way to move an existing FAQ document, support macro list, or sales script into your agent's training without retyping it.
File imports publish straight away: rows that match the expected format become live Q&A pairs as soon as the import finishes, and exact duplicates of pairs you already have are skipped. They don't wait in Needs review, so check the file before you upload it.
You can also inline-edit published answers directly in the Q&A list when you need a quick wording tweak. Empty answers are rejected - every pair needs a real answer. If a save fails, the previous wording rolls back so you don't lose what was already published.
Best practices
Before a big import, scan your file for near-duplicate questions. If the same question appears twice with different answers, the agent has no reliable way to pick the right one. Because imported pairs go live immediately, fix wording in the file first rather than planning to clean it up in review. Write questions the way a visitor actually types them, not the way you'd file them internally - the closer the wording matches real chat traffic, the more reliably the agent will match on it later.
Importing eWebinar Saved Replies
What it is
If your agent is linked to an eWebinar webinar, you can import your existing eWebinar Saved Replies - the canned answers you or your moderators already use in that webinar's live chat - as a starting set of Q&A candidates.
How it works
From the Q&A tab, click Import saved replies. The import starts right away and pulls your library in as candidate pairs in Needs review; greetings and missed-chat messages are skipped because they aren't real answers. Afterwards, Check again picks up any Saved Replies you've added since, and you can dismiss the card if you don't want it. It's the quickest way to give a new agent a head start, since it's answers you've already written for that exact audience and already trust.
Best practices
Import Saved Replies early, right after setup, so your agent has a baseline of trusted answers before it ever talks to a live visitor. Skim the import for replies written for a specific moment in a specific live session - like "we'll get to that at the end" - those need editing or rejecting rather than approving as-is, since your agent might be asked the same question at any point in the video.
Reviewing suggested Q&A
What it is
Suggestions - imported Saved Replies, and new pairs or updates that past conversations suggest - land in a Needs review queue first. Nothing in that queue is used to answer visitors until you approve it, so a suggestion can't quietly change what your agent says. (Conversation-sourced suggestions are still just candidates: the chats are not training, and only the Q&A you approve becomes training.) Suggestions waiting for review don't use training storage either.
How it works
Work through the Needs review queue and approve, edit, or reject each suggestion. If the agent can already answer a suggested question well using its other training, it moves that item to Not needed - so you're not stuck approving pairs that would just duplicate what the agent already knows. You can restore a Not needed item if you want to lock in exact wording anyway.
When a suggestion updates an existing published pair, new question wordings appear in an Adding this question box, and if the answer changed you see the Current answer and Proposed answer with the differences highlighted - you aren't left comparing the old and new wording by eye.
Questions that came up in conversations without a good answer land in Needs answers. Write the answer yourself, or use Check training to have eChat check whether your current training already answers it (or Generate answer when there's a partial answer to build on). Each check uses one credit when it runs.
From Playground (and other test modes), a View in Training data link on a Q&A citation jumps you straight here - pinning a published pair or highlighting one still in Needs review - so you can inspect or fix what the agent grounded on without losing your test conversation. See trying it in Playground.
As your Q&A library fills in, expect more of what lands here to be Conversations suggesting a refinement to a pair you already have, rather than a brand-new Saved Replies import - that's the normal pattern once the basics are covered.
Best practices
Review suggestions in one sitting, soon after they arrive - a large pending queue is easy to forget about, and until you approve a pair, the agent keeps answering that question the old way. When a suggestion is close but not quite right, edit it rather than reject it outright; that's usually less work than writing a fresh pair from scratch, and it still gets you a correct answer in the queue. Use the highlighted differences on published-pair updates to decide quickly whether the new wording is an improvement.